Thanks for the details.
Increasing the CPUs Clock Speed would give you more performance, especially with more complicated models, and 3.2 to 3.6GHz should give a noticeable improvement.
Something else that’s worth validating … When you have time outside of business hours (or during business hours if you have an empty server with the same Specification), try running the same workload but with only 1 user connected to the Server and see if there’s a performance or experience difference. The reason being that your CPUs now have a 4.5x overcommit on them, assuming all 28 users are running 8 vCPUs and there is nothing else running on the Servers, whereas previously is was closer to 2.3x. This would tell you whether you’d be better going for more Cores or increased Clock Speed. For example, with the 3.4GHz 16 Core CPUs mentioned previously, you’d get a slightly faster Clock, and reduce the overcommit to 3.5x if everything else stayed the same. It’s always worth knowing at what point your hardware’s performance starts to deteriorate, and if 2 or 3 people start rendering at the same time or with overlaps on the same Host, you could end up with peaky performance which is really frustrating, this is more likely to happen with increased vCPUs, so it’s worth keeping an eye on, especially if drawings start to become more complex.
On the platforms I design, deliver and support, I always make sure they’re kept well up to date and are running the latest versions of Software and Operating Systems. vGPU drivers are always kept up to date as well. The Applications get treated a little differently, as moving to newer Applications can initially reduce productivity if there are significant differences between that and the previous version. It’s easy for me to just say run the latest Applications, when I’m not the person that will actually be using them. However, the recommendation to my customers is always try and run the latest versions of whichever Applications you use to take advantage of the latest features, performance and functionality. However, the users may (and probably will) need some time to adapt before any real performance metrics can be gathered. So to directly answer your question, if the users are happy to do so, then I would be running AutoCAD 2020 over 2019.
For the NVIDIA Control Panel, I typically let the users decide what they want to tweak inside there. This is always going to be Application specific as to any benefits and as AutoCAD is typically CPU limited, modifying the NVIDIA Control Panel doesn’t always give much benefit in my experience. There are some basic settings inside AutoCAD that can make things run more smoothly, and there are plenty of random guides on the internet of things to try that relate to the entire Operating System. For example, have you run the Citrix or VMware Optimiser tools?
If you’re not locked into those server specifications and this is still a POC, I think the biggest gain you’re going to see will be CPU related by going for a different model. If these are your production servers, then you’re effectively trying to turn things off to improve existing performance, so system wide optimisations are key, which is why I mention the Optimisation Tools.